1

Department of Economics

Issn 1441-5429

Discussion paper 09/10

Long-Run Convergence in Manufacturing and Innovation-Based Models1

Jakob B. Madsen2 and Isfaaq Timol3

Abstract: Most studies of comparative productivities fail to find evidence of convergence in OECD

manufacturing despite major economic growth theories predicting convergence. Using

manufacturing data for 19 OECD countries over the period from 1870 to 2006 this study finds strong

evidence of unconditional -convergence as well as -convergence. Panel data estimates suggest

that the convergence has been driven by domestic R&D, international R&D spillovers and financial

development as predicted by Schumpeterian growth theories.

JEL Classification: E13, E22, E23, O11, O3, O47.

Key words: Convergence, second-generation endogenous growth models.

1 Helpful comments and suggestions from Steve Dowrick, Mark Harris, Don Poskitt, seminar participants at Melbourne University and

University of Western Australia and especially two referees, are gratefully acknowledged. Jakob B Madsen acknowledges financial

support from an Australian Research Council Discovery Grant No DP0877427. 2 Department of Economics, Monash University 3 Department of Econometrics and Business Statistics, Monash University

© 2010 Jakob B. Madsen and Isfaaq Timol

All rights reserved. No part of this paper may be reproduced in any form, or stored in a retrieval system, without the prior written

permission of the author.

2

1. Introduction

Following the papers of Broadberry (1993), Wolf (1994), Bernard and Jones (1996a, 1996b) and

Carree et al. (2000), it is widely believed that labor productivity in manufacturing has not been

converging across the OECD countries.4 Bernard and Jones (1996a) conclude that the service sector

has been the driving force behind the finding of economy-wide convergence in the literature.5

However, since trade in services has been low compared to trade in intermediate manufacturing

goods, we would expect any international transmission of technology to be stronger in manufacturing

than in services, to the extent that convergence is driven by trade. Therefore, we should expect

convergence in manufacturing to have been more pronounced than convergence in services. Ben-

David (1993), for example, finds that the movement towards freer trade over the past century has

been an important factor behind the per capita income convergence. This argument is supported by

Baumol (1986) who argues that expansion in exports over the period 1870 to 1979 amplified

international competition and, consequently, was conducive to imitation and innovation.

Furthermore, the “advantages of backwardness” along the lines of Gerschenkron (1962) would

suggest catching up to the frontier in all sectors of the economy (Dowrick and Gemmell, 1991,

Bernard and Jones, 1996a).

The convergence tests in the 1990s were often used to discriminate between first-generation

endogenous growth theories and neoclassical growth theories under the assumption that endogenous

growth models do not predict convergence (see for example Mankiw et al., 1992). However, only a

very few of the early first-generation endogenous growth models do not predict convergence. Kelly

(1992), for example, showed that convergence tends to occur in early first-generation endogenous

growth models when stochastic factor productivity is introduced. More importantly, endogenous

growth models have come a long way since then and have increasingly focused on the role of

technology transfer and absorptive capacity in explaining productivity growth across countries

(Eaton and Kortum, 1999, Howitt, 2000, Griffith et al., 2003, 2004, Aghion, Howitt and Mayer-

Foulkes, 2004, 2005, Madsen, 2008a,b). In the Schumpeterian models of Aghion and Howitt (2005),

and Aghion et al., (2004, 2005), countries with highly productive R&D, adequate property right

protection and good educational systems will converge. Furthermore, Madsen (2008b) finds that

growth can be permanently affected by knowledge spillovers. This puts the convergence debate

today into quite a different light from that of the 1990s.

4 Edward Wolff (1991) and Dollar and Edward Wolff (1988) do find convergence in manufacturing. However, Bernard

and Jones (1996b) argue that there are problems associated with the data used by Dollar and Wolff (1988). 5 For findings of economy-wide convergence among the OECD countries, see, for example, Baumol (1986), Baumol and

Wolff (1988), Dowrick and Nguyen (1989), Sala-i-Martin (1996), and Madsen (2007).

3

Distance to the frontier also plays a particularly important role in the convergence debate.

Countries that are more backward technologically may have greater potential for generating rapid

growth than more advanced countries (Gerschenkron, 1962), essentially because backwardness

reduces the costs of creating new and better products (Howitt, 2000). However, backwardness needs

not automatically lead to growth since the increasing complexity of products requires large

investments in knowledge in order to take advantage of the technology developed elsewhere (Aghion

et al., 2004, 2005). Large investments in R&D require a developed financial system that can provide

inventors with sufficient capital to finance their R&D expenses (Aghion et al., 2004, 2005) and

factory workers, technicians, engineers, and managers need to be trained to use technologies

developed elsewhere (Hobday, 2003).

Taking into account the recent developments in endogenous growth theories, this paper tests

for conditional and unconditional convergence in OECD manufacturing. The contribution of the

paper is two-fold. First, it considers a substantially longer data period than has previously been used

in producing empirical estimates for a large sample of OECD countries. Second, it tests the extent to

which convergence has been driven by R&D, knowledge spillovers, human capital, financial

development and the interaction between distance to frontier and human capital, research intensity

and financial development, following the prediction of second-generation models of economic

growth. Using a new dataset for the manufacturing sector covering up to 137 years for 19 OECD

countries this paper tests 1) whether manufacturing total factor productivity (TFP) and labor

productivity have converged over time; and 2) whether R&D, human capital, international

knowledge spillovers through the channel of imports, the distance to the frontier and the interaction

between the distance to the frontier and financial development, human capital and research intensity

have contributed to productivity convergence or divergence in manufacturing.

The country sample used in the paper satisfies two important criteria. First, that the countries

have good legal systems, a high quality educational system, and developed credit markets (Aghion

and Howitt, 2005, Aghion et al., 2004, 2005). Second, that the sample includes countries that were

well behind the technology frontier during the 19th

and a significant part of the 20th

century including

Ireland, Japan, Portugal and Spain. Thus, the country sample, to a large extent, overcomes De Long‟s

(1988) critique of country selection bias in Baumol‟s (1986) study of per capita GDP convergence

among the industrialized countries since 1870. De Long‟s (1988) main concern was that most papers

on long-term convergence consisted of countries that were already well developed in the 20th

century. Consequently, their results were biased towards the finding of convergence since countries

that were likely to diverge in the twentieth century such as Argentina, Ireland, Portugal and Spain,

were left out of the sample.

4

The next section discusses the convergence predictions of various growth theories including

recent second-generation endogenous growth models. Section 3 provides graphical evidence and

tests of unconditional convergence, while Section 4 tests for conditional and unconditional

convergence using a panel data approach. Section 5 concludes.

2. Convergence in endogenous growth models

Since the publication of Mankiw et al. (1992) it has been widely believed that first-generation

endogenous growth models do not predict productivity convergence. However, the endogenous

growth models referred to by Mankiw et al. (1992) were the simple “AK” type models, which were

only used in a few early endogenous growth models and, as such, are unrepresentative of first-

generation endogenous growth models. The first-generation models of Lucas (1988) and Romer

(1990), for example, exhibit conditional convergence and each country converges to its own steady-

state growth rate. Due to the unwarranted property of proportionality between productivity growth

and the number of R&D workers in first-generation endogenous growth models, they have been

replaced by second-generation endogenous growth models; namely semi-endogenous growth models

and Schumpeterian growth models.

The semi-endogenous growth models by Jones (1995, 2002) and Kortum (1997) avoid scale

effects and assume decreasing returns to knowledge stock. In the Schumpeterian growth models of

Peretto (1996, 1998, 1999b), Aghion and Howitt (1998), Dinopoulos and Thompson (1998), Howitt

(1999, 2000), and Peretto and Smulders (2002), R&D has to increase over time to keep economies

growing at constant rates. This is because the increasing range of products as the economy expands

lowers the productivity effects of R&D activity. Schumpeterian models dispose of the scale effects

by the assumption that innovations occur at the firm level instead of at the economy-wide level. In

other words, Schumpeterian theory shifts the focus from the whole economy to the individual

product line under the assumption that there is one product line per firm.

What do the second-generation growth models say about convergence? Semi-endogenous

growth models possess the same steady state properties as the Solow model and, as such, predict

conditional convergence (see for example Jones, 2002). Since growth is temporarily affected by

growth in R&D and human capital, the transitional dynamics will be different from that of the Solow

model. Jones (2002) shows that the transitional dynamics are slower in semi-endogenous growth

models than in the Solow-Swan model because of the interaction between fixed capital and

knowledge. The Schumpeterian models developed by Peretto (1998, 1999a,b, 2003), Howitt (1999,

2000), Aghion et al. (2004, 2005), Howitt and Mayer-Foulkes (2005), and Aghion et al. (2006) also

predict conditional convergence. To see this consider the following model of Aghion and Howitt

5

(2005) in which there is a theoretical mechanism that drives equalization of growth rates in the long

run. As long as a country innovates it will eventually start growing at the same rate as the leading

countries. On the other hand, countries with poor macroeconomic conditions, institutions,

educational systems and underdeveloped financial systems will stagnate.

Aghion and Howitt (2005) demonstrate that country i‟s expected distance to the technological

frontier, , evolves according to the equation:

, (1)

where i is country i‟s innovation rate, is the global innovation rate, and is the size of

innovations. Country i‟s innovation rate is given by , where n is productivity-adjusted

research, f(n) is the research productivity function and is R&D productivity. The distance from the

frontier at time t-1 is given by , where A is a productivity parameter and is

frontier technology.

If > 0, this differential equation is stable, which means as long as a country undertakes

R&D at a constant intensity, n, its distance to the frontier will stabilize at zero and its growth rate

will converge at the same rate as the growth rate at the technology frontier. If = 0, there is no stable

equilibrium and diverges to infinity: the country stops innovating and will, therefore, have a long

run productivity growth rate of zero.

This framework shows that countries either fall into a group in which they converge to the

frontier growth rate (i.e. > 0) or a low income group (i.e. = 0). The high income group consists of

countries with highly productive R&D, a good educational system, and good property right

protection. These countries will converge to the frontier growth rate (Howitt, 2000, Aghion et al.,

2004, 2005). Countries with low R&D productivity, poor educational system, and low property

rights will not grow at all. The countries considered in this paper have had appropriate institutions in

most of the period 1870-2006 (see Jaggers and Marshall, 2007), at least some basic education at the

turn of the 20th

century (Bayer et al., 2006), and have undertaken R&D throughout the whole period

(Madsen, 2008a). Accordingly, these countries should converge.

More precisely, Aghion and Howitt (2005) show that a country undertakes R&D and catches

up to the frontier if the marginal benefits from R&D exceed the marginal cost:

, (2)

6

where is the price mark-up over marginal costs, which depends on property right protection, L is

the supply of skilled labor, and m is the number of sectors in the economy. A country is more likely

to undertake R&D and converge to the growth rate of the frontier country the better is the

educational system, as measured by λ, the better is the protection of property rights, as measured by

χ, and the more skilled is the labor force, as measured by L. Thus, in this model there is a clear

mechanism that drives equalization of growth rates in the long run. This is a desirable property that is

not shared by closed-economy growth models.

Aghion et al. (2004, 2005) and Aghion and Howitt (2005) extend this framework to allow for

financial development. Financial development is important for convergence because it determines

the degree to which borrowers choose to defraud creditors by concealing the profits of the R&D

project in the event of success. Aghion et al. (2004, 2005) show that the more financially developed

a country is, the more difficult it is to defraud creditors and the easier is the access to credit to

undertake R&D. If credit markets are functioning perfectly Equation (2), modified with a one-period

discount factor, will still hold. If credit markets, however, are imperfect, investment is limited by a

fixed multiple of accumulated net wealth, which in turn, constitutes current per capita income. It

follows that the further a country falls behind the frontier country the less the entrepreneur will be

able invest in the R&D that is required to maintain a given frequency of innovations. If, on the other

hand, the costs of defrauding are sufficiently high, even a very backward country can take advantage

of its backwardness in the domain of the frequency of innovations. These considerations suggest that

financial development plays a potentially important role for convergence, an issue that is examined

in Section 4.

3. Unconditional convergence

3.1 Data

The country sample consists of the following 19 OECD countries: Australia, Belgium, Canada,

Denmark, Finland, France, Germany, Ireland, Italy, Japan, Netherlands, New Zealand, Norway,

Portugal, Spain, Sweden, Switzerland, United Kingdom, and the United States. Productivity is

measured as manufacturing labor productivity as well as TFP. The labor productivity data cover the

period 1870 to 2006 while the TFP data cover the slightly shorter period 1900 to 2006 because data

on manufacturing investment are only available for a few countries before 1900. The labor

productivity data has the advantage over the TFP data in that it spans 30 years further back, while the

TFP data has the advantage of catering for the feed-back effects from capital accumulation in the

convergence regressions. Suppose that convergence is driven by capital accumulation as predicted by

7

the neoclassical model. If labor productivity is regressed on R&D, distance to the frontier and other

variables in the convergence regressions, one may come to the conclusion that the convergence is

driven by R&D while, in fact, it is driven by capital accumulation because of a potentially high

correlation between capital accumulation and the R&D variables.6 TFP regressions ensure that the

conditional variables considered below influence productivity through technological progress and not

through capital accumulation.

In contrast to the majority of studies on manufacturing productivity convergence, labor

productivity and TFP are based on hours worked as opposed to number of workers. Bernard and

Jones (1996b) claim that they are the first to allow for hours worked. Adjustment for annual hours

worked is particularly important in this study because annual hours worked has been reduced to a

half over the past 137 years and because the cross-country variation of hours worked has converged

among the countries considered here, as shown below. Labor productivity is measured by real

manufacturing GDP in 2002 purchasing power parities (PPP) divided by manufacturing employment

and the average annual hours actually worked per person in the non-agricultural sector. Hours

worked in the non-agricultural sector is likely to be a good proxy for manufacturing hours worked

since most of the changes in hours worked over time have been driven by the number of public

holidays and regulations regarding number of weekly hours worked.

The TFP estimates are based on the Cobb-Douglas production function, Y = Kα(AL)

1-α, where

Y is manufacturing output, K is manufacturing capital stock, A is the level of technology, and L is

total employment in manufacturing times annual hours worked. Harrod-neutral technological

progress is assumed to make the steady-state TFP growth rates comparable with the steady-state

labor productivity growth rates. Here, A can be straightforwardly computed as:

.

Capital stock is calculated from manufacturing investment using the perpetual inventory method and

a depreciation rate of eight percent. Capital stock data are available for 11 countries in 1900 and

gradually become available for the other countries after this period. The capital stock is available for

all countries from 1950 except Australia and Switzerland. TFP is backdated using labor productivity

in the periods for which capital stock is not available (see data appendix for details). Capital‟s

income share, α, it set to 0.3 following the standard in the literature (see for instance Mankiw et al.,

1992, Jones, 2002, Madsen, 2008b). We have not allowed the income share of capital to vary over

6 We are grateful to a referee for pointing this out.

8

time because only recent data on income shares are available and variation in income shares are more

likely to reflect variations in rent extraction than changes in marginal productivities of production

factors (see for example Bruno and Sachs, 1985, and Hall, 1988). Furthermore, Gollin (2002) shows

that variations in factor shares across countries, particularly, and over time are heavily influenced by

the rate of self employment. Earnings from self employment are recorded as profit income in

national accounts although the labor of the self employed should be attributed labor income.

Correcting for imputed labor income of the self employed for a large sample of countries Gollin

(2002) finds that income shares are quite constant across countries.7

Before WWII the manufacturing value added production data are mostly based on

manufacturing or industrial production figures obtained from surveys of establishments or tax files,

while the employment data are often based on census surveys. Since alternative sources for

manufacturing production and employment are not available before WWII, except in a couple of

instances, the quality of our data cannot be checked against other sources. Regarding annual hours

worked, the analysis by Madsen et al. (2010) suggests that the annual hours worked used in this

study are at least as good as those of alternative sources. Although the manufacturing productivity

data far back in time is not of the same quality as the manufacturing data available today, the data are

probably of much better quality than the economy-wide productivity data, which have been used in

most other convergence studies. The problem associated with economy-wide GDP data is that GDP

cannot be measured adequately in several sectors of the economy including government and most

private services including health, banking and insurance, defence, and space (Griliches, 1979).

Furthermore, historical economy-wide GDP estimates are often interpolated, aggregated over

incomplete sectoral data or expenditure components, and based on indirect indicators. Manufacturing

GDP data do not suffer from the same deficiencies and, as such, can give more reliable estimates of

productivity than economy-wide estimates.

3.2 Graphical analysis

Figures 1 and 2 show the evolution of the log of labor productivity in the period 1870-2006, and the

log of TFP in the period 1900-2006. Both graphs suggest convergence since the gap between the

most productive and least productive countries has been decreasing over time. The indication of

negative cross-sectional correlation between initial productivity and subsequent growth rate suggests

-convergence. The US, the UK and Switzerland have been the countries with the highest labor

7 Peretto and Seater (2008) show that technological progress endogenously reduces the output elasticity of the non-

reproducible factors of production such as land and natural resources. Since land is not an important factor of production

in manufacturing our estimates are unlikely to be influenced by the Peretto-Seater effect.

9

productivities during most of the period. In terms of TFP, Japan took over as the most productive

nation after 1970. Portugal has had the lowest labor and total factor productivities during the whole

sample period and ceased to converge to the mean over the past three decades. This poor

performance over the past three decades has also been observed by the OECD (2004), which

attributes the low growth to inefficient allocation of capital equipment in the business sector, late

adoption of new technologies, low levels of education compared to other OECD countries, poor

access to training and an unattractive business environment.

10

Since -convergence is a necessary but not a sufficient condition for convergence (Sala-i-Martin,

1996) it is also necessary to consider -convergence. Figure 3 shows marked -convergence. The

standard deviation in 1998 was a quarter of that in 1870 for labor productivity and a third of that in

1900 for TFP. The convergence is concentrated in the 20th

century. The slight divergence since 1998

is due to the Irish productivity boom that pushed Ireland ahead of the other countries, combined with

Portugal falling further behind. Overall, there seems to be clear evidence of -convergence in labor

productivity over the past 137 years and in TFP over the past 107 years. Finally, Figure 4 shows a

clear decline in the variation of hours worked in our country sample, with a significant fall

immediately after WWII. This fall was predominantly driven by a marked reduction in hours worked

in Japan and Germany towards the mean. In total, there has been a 75% reduction in the cross

country standard deviation in annual hours worked in the period 1870-2006.

Note. The standard deviation is based on the log of productivity and the levels of annual hours worked.

3.3 Tests of unconditional convergence

This section tests for unconditional -convergence as well as -convergence. Testing for

convergence involves the following two regressions:

, i = 1, 2,…, 19, (3)

and

, i = 1, 2,…, 19, (4)

where is the average labor productivity growth rate in the period 1870-2006, is

the average TFP growth rate in the period 1900-2006, , , and are constants, and

11

are the initial productivities for country i, and is a stochastic error term. Here, and ‟

determine the relationship between initial productivities and subsequent growth rates.

Table 1. Parameter estimates of Eqs. (3) and (4).

0.074 0.099

(0.000) (0.000)

-0.007 -0.010

(0.000) (0.000)

0.83 0.80

Notes. The numbers in parentheses are p-values. The estimation period is

1870-2006 for labor productivity and 1900-2006 for TFP.

The results of regressing Eqs. (3) and (4) are shown in Table 1. Since the estimated coefficients of

and ‟ are negative and highly significant at conventional levels of significance, the null hypothesis

of no -convergence is easily rejected. This confirms the graphical evidence above that countries

with high levels of productivity in 1870 or 1900 have been growing at slower rates during the period

1870-2006 or 1900-2006 than countries with lower initial productivity levels.

The tests developed by Lichtenberg (1994) and Carree and Klomp (1997) are used to test for

-convergence. The test statistic of Lichtenberg (1994) is constructed as and has an F

distribution with ( degrees of freedom in both the numerator and the denominator. Here is

the cross-country variance of labor productivity in the first period, T0, (1870 or 1900), is the

variance in the last period, T, (2006), and N is the number of countries in our sample. The likelihood

ratio test of Carree and Klomp (1997) is constructed as follows:

,

where is the productivity covariance between period T and T0. The test statistic is distributed as

under the null hypothesis of no convergence. The results from these two tests are presented in

Table 2. Both tests give evidence of -convergence in manufacturing labor productivity and TFP at

the 1-percentage significance level.

Table 2. Sigma convergence tests.

12

Lichtenberg's test 5.57 3.86

[3.24] [3.24]

Likelihood-ratio test 11.12 7.02

[6.63] [6.63]

Note. The numbers in square brackets are critical values at the one percent

significance level.

3.4 Robustness tests of unconditional convergence

This sub-section investigates the robustness of the convergence tests to 1) the exclusion of

„problematic‟ countries; 2) the sample used by Bernard and Jones (1996b); and 3) different PPP base

years. The first rows (Case 1) in Table 3 address De Long‟s (1988) sample selection issue by

examining whether our results are sensitive to the exclusion of the four „problematic‟ countries

mentioned in his paper, namely, Ireland, Portugal, Spain and New Zealand. Excluding these

countries from our sample has negligible effects on the results obtained in the previous sub-section.

There is strong evidence of -convergence as well as -convergence when the four „problematic‟

countries are excluded.

Table 3. Robustness tests of unconditional convergence.

Dep. Var. β

Lichtenberg's

test

Likelihood-ratio

test

---------------------------------------CASE 1---------------------------------------

-0.007 14.283 17.902

(0.000) [3.905] [6.635]

-0.010 8.052 11.829

(0.000) [3.905] [6.635]

---------------------------------------CASE 2---------------------------------------

-0.029 2.500 3.722

(0.009) [4.155] [6.635]

-0.020 1.543 1.164

(0.025) [4.155] [6.635]

---------------------------------------CASE 3---------------------------------------

Our Y/Emp from B&J -0.024 1.602 1.102

(0.018) [4.155] [6.635]

Y from B&J/ Our Emp -0.032 1.251 0.179

(0.027) [4.155] [6.635]

Our Y/ Emp from ILO -0.027 1.596 0.961

(0.016) [4.155] [6.635]

Our data and B&J‟s PPP -0.034 1.851 1.398

(0.005) [4.155] [6.635]

---------------------------------------CASE 4---------------------------------------

using 1990 PPP -0.007 7.193 14.009

(0.000) [3.242] [6.635]

using 1980 PPP -0.008 6.254 12.307 (0.000) [3.242] [6.635]

using 1990 PPP -0.010 4.527 8.760 (0.000) [3.242] [6.635]

13

using 1980 PPP -0.010 4.266 8.086 (0.000) [3.242] [6.635]

Notes. The numbers in round parentheses are p-values while critical values at the one percent level are in

square brackets. Case 1. The four „problematic‟ countries (Ireland, Portugal, Spain and New Zealand) are

excluded from the sample. Case 2. The sample period and country sample of Bernard and Jones (1996b)

(B&J) is used. Case 3. The employment data of Bernard and Jones (1996b) and our income data are used in

the first row of this case. The income data of Bernard and Jones and our employment data are used in the

second row of this case. Our income data and ILO‟s employment data are used in the third row of this case.

Our data converted to PPP by B&J‟s PPP in the fourth row of this case. Productivity is measured as labor

productivity in Case 3. Case 4. 1980 and 1990 PPPs (instead of 2002 PPP in the baseline case) are used as

conversion factors.

Case 2 considers the country sample and time period of Bernard and Jones (1996b) using our

productivity data (their sample consists of the following 14 countries in the period 1970-87:

Australia, Belgium, Canada, Denmark, Finland, France, Germany, Italy, Japan, Netherlands,

Norway, Sweden, the UK, and the US). The estimated coefficients of -convergence are quite close

to the ones obtained by Bernard and Jones (1996b); however, in contrast to the finding of Bernard

and Jones (1996b), the null hypothesis of no -convergence is rejected at conventional significance

levels.

This raises the question as to why the null hypothesis of no -convergence is rejected in our

sample but not in Bernard and Jones‟s. To investigate this issue Case 3 in Table 3 considers 1) our

income but their employment data; 2) our employment but their income data; 3) ILO‟s employment

data as opposed to the OECD employment data (which are used by Bernard and Jones as well as in

this paper); and 4) their PPP conversion values. In all these instances there is still evidence of -

convergence but not -convergence. These results suggest that the conflicting results between this

paper and Bernard and Jones reflect revision of the income and employment data. Data are often

revised several years back in time and can sometimes result in significant changes. Comparing their

data with ours reveals that the discrepancy is quite small. In this context it is important to note that

there are only small differences between their and our results: the null hypothesis of no -

convergence cannot be rejected in our as well as in their case during the period 1970-87, there is no

significant difference between the estimated coefficients at conventional significance levels, and

the null of no -convergence is even rejected by Bernard and Jones if a one-sided 10-percentage

benchmark significance level is applied.8 The discrepancy is, therefore, small and the

8 Bernard and Jones (1996b) say that they find no convergence at the 10 percentage level. This conclusion is based on a

two-sided critical value. We base our conclusion on a one-sided critical value because we test whether < 0 and not

whether is significantly different from zero.

14

inconclusiveness ultimately rests on the very short and volatile data period chosen by Bernard and

Jones. Had a longer sample period been chosen, the results would, in all likelihood not have been

sensitive to later revisions of the data.

Finally, in his comments on the paper by Bernard and Jones (1996b), Sørensen (2001) argues

that, whether a particular sample of countries exhibits productivity convergence depends on the

choice of base year. Extending the sample used by Bernard and Jones by six years, Sørensen (2001)

finds that the earlier the base year, the lesser is the evidence of productivity convergence in

manufacturing. To investigate this issue, we check the robustness of our results using 1980 and 1990

PPPs (instead of 2002 PPP) as conversion factors (Case 4). The test results in Table 3 show

significant evidence of -convergence as well as -convergence. The null hypothesis is strongly

rejected in all cases. In conclusion, our results seem to be invariant to the choice of PPP base year,

providing a higher degree of confidence of - and -convergence in manufacturing labor

productivity and TFP.

4. Panel estimates of convergence

The finding of unconditional convergence raises the question of which factors have been responsible

for the convergence. Panel estimates are undertaken in this section to examine whether innovation

based variables, human capital, and distance to the frontier can account for convergence and

manufacturing productivity growth. Restricted and unrestricted versions of the following model are

estimated:

(5)

where the superscripts d and f stand for foreign and domestic, is labor productivity for country i

in period t, is labor productivity at the start of each period over which the long differences are

taken, H is educational attainment (average years of schooling among the adult population), Pat is

the number of patent applications, Emp is economy-wide employment, is country dummies, and

DWWII is a dummy variable taking the value 1 before WWII (1950) and 0 afterwards, which is

included to capture the increasing productivity growth in the post 1950-period that may not be

accounted for by the explanatory variables. Finally, measures the distance to the

technological frontier at the beginning of each period and is measured as ln(yUS

/yi), where yUS

is

15

manufacturing productivity in the US. Equation (5) is also regressed using TFP instead of y. In these

regressions the initial productivity and the distance to the frontier are also measured in terms of TFP.

The model is estimated in 5, 10 and 15 year differences to filter out business cycle influences.

All variables are measured as average annualized growth rates and research intensities, (Pat/Emp),

are measured as the average in the period over which the first-differences span. The innovative

activity is measured by the total number of patents applied for, since patents are the only presently

available data on innovative activity dating back to 1870. Patents are normalized by economy-wide

employment and not manufacturing employment because the number of patents covers the whole

economy. Economy-wide patents are used because industrial patents are only available for some

countries and mostly cover a short time-span.

The model is deliberately made as inclusive as possible 1) to prevent omitted variable biases;

2) to test for the possibility that knowledge can have both permanent and temporary growth effects

following the predictions of second-generation endogenous growth models; and 3) to ensure that as

many factors as possible that can potentially explain growth and convergence are included in the

model. The predictions of the two leading second generation growth models, namely semi-

endogenous growth theory and Schumpeterian growth theory, are allowed for in the estimates. While

the semi-endogenous growth theory by Jones (1995) abandons scale effects in ideas production, the

Schumpeterian growth models of Aghion and Howitt (1998), Peretto (1996, 1998), Howitt (1999,

2000) and Peretto and Smulders (2002) maintain scale effects but assume that the effectiveness of

R&D dilutes due to the proliferation of products as the economy expands.

The two second generation models have quite different implications for growth. As shown by

Laincz and Peretto (2006) and Madsen (2008b), Schumpeterian theory predicts that labor

productivity is growing proportionally with research intensity, which is measured as patents divided

by employment in the estimates of Madsen (2008b). Patents are divided by employment to allow for

product proliferation and increasing complexity of new innovations as productivity increases (Ha and

Howitt, 2007). Similarly, Peretto (1999b) shows that an employment-induced increase in firms‟

profit rates brings the growth rate temporarily up to a higher level because the incumbents invest

more in R&D. The higher rate of profit induces an entry of new firms, which in turn attracts

employees of the incumbents. This process continues until the rate of profit and the productivity

growth rates slow down and revert towards their original steady state levels.

Growth can still be sustained at a constant rate in the Schumpeterian framework if R&D is

kept in a fixed proportion of the number of product lines, which is in turn proportional to the size of

population in steady state. As such, to ensure sustained TFP growth, R&D has to increase over time

to counteract the increasing range and complexity of products that lowers the productivity effects of

16

R&D activity. Similarly, the Schumpeterian model of Aghion et al. (2006) predicts that TFP growth

is proportional to educational attainment, which implies that the growth rate will remain positive as

long as the labor force has some education.

Semi-endogenous growth theory, by contrast, assumes that educational attainment and R&D

have only temporary growth effects (see for example Jones, 2002). Accordingly, productivity growth

rates are positively related to growth in R&D and the change in educational attainment. Positive

productivity growth is only feasible as long as educational attainment and R&D are growing at

positive rates. In steady state this means that population growth rates have to be positive to get

positive productivity growth rates (Laincz and Peretto, 2006).

The growth in foreign patents and foreign research intensity affect growth following Coe and

Helpman (1995) and Madsen (2007, 2008a, 2008b), in which productivity growth is affected by

knowledge spillovers through the channel of imports. The idea behind this spillover hypothesis is

that the variety and the quality of intermediate inputs are predominantly explained by R&D and,

therefore, productivity is a positive function of R&D. Consequently, the productivity of a country

depends on its own R&D and the R&D embodied in imported intermediate inputs and, therefore, that

technology is transmitted internationally by import-weighted R&D. Here, imports of technology are

allowed to follow the semi-endogenous growth hypothesis through the growth in foreign patents, and

in the Schumpeterian growth theory through foreign research intensity. 9

Knowledge spillovers

through the channel of imports are not only important because they play an important role for growth

in endogenous growth models but also because trade has often been highlighted as playing a key role

in facilitating convergence (see for example Nelson and Wright, 1992, and Ben-David, 1993).

The DTF term captures the idea that there are benefits to backwardness, following the

historical analysis of Gerschenkron (Howitt, 2000) and the empirical analysis of Dowrick and

Gemmel (1991). The distance to the frontier also impacts on productivity by interacting with

research intensity. In the Schumpeterian model of Howitt (2000), a country takes advantage of its

9 Imports of patents through the channel of trade of country i, Pat

f, are based on the following weighting schedules

suggested by Lichtenberg and Van Pottelsberghe de la Potterie (1998):

djt

jnjt

ijtfit Pat

Y

MPat

21

1

, ji . Semi-endogenous

d

jtjnjt

ijtf

itEmp

Pat

Y

M

Emp

Pat

21

1

, ji . Schumpeterian

where Mij is nominal imports of goods from country j to country i, and n

jY is nominal income of country j.

17

backwardness directly through the distance to the frontier and indirectly through the interaction

between the absorptive capacity and the distance to the frontier. In Howitt‟s (2000) model, it is R&D

intensity that draws a country to the technology frontier and the higher is the research intensity, the

faster the country converges to the technology frontier. The interaction between DTF and educational

attainment is included in the estimates to allow for the possibility that education enhances the

absorptive capacity of a country following the hypotheses of Nelson and Phelps (1966) and Howitt

and Mayer-Foulkes (2005).

4.1 Estimation method

The model is regressed using the corrected least squares dummy variable (LSDV) of Kiviet (1995).

The appendix reports the results when alternative estimators are used. The results remain almost

unaltered using these estimators. The LSDV estimator is bias-corrected as parameter estimates using

the traditional uncorrected LSDV estimator can be substantially biased in samples with small T’s like

ours (in our sample T = 27, T = 14, and T = 9 when using 5-year, 10-year, and 15-year intervals

respectively). Based on Monte Carlo simulations, Kiviet (1995) finds that this corrected estimator is

very accurate for small values of N and T and is more efficient than several IV estimators.

4.2 Panel tests of unconditional convergence

Before regressing Eq. (5) we examine whether the findings of unconditional convergence from the

previous sections can be maintained using the panel approach. While we tested for convergence in

2006 relative to 1870 or 1900 in the previous estimates, the panel approach tests for -convergence

in period t relative to the periods t-5, t-10 and t-15, where t is measured in years. The estimation

results are reported in Table 4. The estimated coefficients of initial productivity, , are consistently

negative and highly significant, which reinforces the finding of unconditional convergence in the

previous sections. The speed of adjustment is higher in the TFP estimates than in the labor

productivity estimates, which may reflect that the convergence speed is watered down in the labor

productivity regressions by the period 1870-1900, during which convergence was absent.

Table 4. Tests of unconditional convergence.

5 years 10 years 15 years

-0.119 -0.141 0.117 0.146 0.078 0.073

(0.011) (0.006) (0.073) (0.056) (0.366) (0.478)

-0.010 -0.022 -0.009 -0.016 -0.015 -0.033

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

18

-0.044 -0.056 -0.035 -0.040 -0.049 -0.079

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

Note: The numbers in parentheses are p-values.

4.3 Conditional convergence

Unrestricted and restricted estimates of Eq. (5) are displayed in Tables 5-7, where the coefficients

that are restricted to zero in the restricted model using the general-to-specific procedure with the

five-percent benchmark level (the lagged dependent variables and the initial productivity variables

are maintained in the restricted regressions). The estimated coefficients of initial productivity are

consistently statistically insignificant, even at the five percentage significance level, regardless of

whether the estimates are in 5, 10 or 15 year differences and whether or not the models are restricted.

This result is very important because it shows that the manufacturing productivity convergence is

driven by the conditional variables included in Eq. (5). This is also a strong result because the

estimated coefficients of initial productivity were extraordinarily significant in the unconditional

regressions in Table 4. As such, powerful conditional variables are required to render the initial

productivity level insignificant.

Table 5. Estimates of Eq. (5) in 5-year intervals

Labor Productivity TFP

5 years Full

Model

Restricted

Model Full Model

Restricted

Model

-0.086 -0.087 -0.062 -0.074

(0.073) (0.056) (0.251) (0.181)

-0.008 -0.004 -0.008 0.002

(0.356) (0.413) (0.527) (0.815)

0.028 0.025 0.016 0.020

(0.000) (0.000) (0.247) (0.132)

0.004 0.005 0.003 0.003

(0.228) (0.168) (0.520) (0.507)

0.008 0.011 0.019 0.017

(0.016) (0.000) (0.001) (0.001)

0.005 0.005 0.006 0.007

(0.032) (0.010) (0.051) (0.018)

0.205 0.196 0.240 0.210

(0.019) (0.026) (0.030) (0.043)

-0.035 -0.034 -0.031 -0.030

(0.000) (0.000) (0.004) (0.005)

0.045 0.024 0.032 0.046

(0.055) (0.000) (0.361) (0.000)

-0.015 0.005

(0.382) (0.829)

0.005 -0.004

19

(0.061) (0.517)

0.016 0.030 (0.534) (0.331)

Note: The numbers in parentheses are p-values.

Table 6. Estimates of Eq. (5) in 10-year intervals

Labor Productivity TFP

10 years Full

Model

Restricted

Model Full Model

Restricted

Model

0.133 0.131 0.232 0.205

(0.047) (0.042) (0.000) (0.002)

-0.006 -0.002 -0.010 0.003

(0.431) (0.658) (0.468) (0.704)

0.009 0.009 0.000 0.004

(0.016) (0.006) (0.968) (0.532)

0.002 0.002 0.001 0.001

(0.412) (0.360) (0.826) (0.823)

0.008 0.009 0.016 0.011

(0.007) (0.000) (0.004) (0.035)

0.004 0.004 0.004 0.006

(0.034) (0.013) (0.149) (0.051)

0.070 0.060 0.051 0.028

(0.045) (0.093) (0.409) (0.624)

-0.027 -0.026 -0.023 -0.021

(0.000) (0.000) (0.017) (0.020)

0.032 0.022 0.007 0.036

(0.152) (0.001) (0.833) (0.002)

-0.010 0.015

(0.532) (0.488)

0.002 -0.006

(0.598) (0.127)

0.013 0.039 (0.523) (0.251)

Note: The numbers in parentheses are p-values.

Table 7. Estimates of Eq. (5) in 15-year intervals

Labor Productivity TFP

15 years Full Model Restricted

Model Full Model

Restricted

Model

0.148 0.119 0.270 0.211

(0.054) (0.124) (0.005) (0.035)

-0.019 0.001 -0.019 0.046

(0.104) (0.921) (0.737) (0.057)

0.007 0.010 -0.003 0.003

(0.082) (0.020) (0.745) (0.788)

0.003 0.003 0.005 0.006

(0.101) (0.096) (0.193) (0.113)

0.012 0.012 0.036 0.030

(0.004) (0.000) (0.000) (0.002)

20

0.024 0.018 0.061 0.060

(0.076) (0.214) (0.120) (0.142)

0.104 0.068 0.136 0.074

(0.024) (0.099) (0.125) (0.374)

-0.033 -0.030 -0.021 0.000

(0.000) (0.000) (0.543) (0.996)

0.038 0.031 0.026 0.120

(0.178) (0.000) (0.661) (0.000)

-0.015 0.026

(0.502) (0.520)

0.002 -0.016

(0.616) (0.052)

0.060 0.148 (0.082) (0.143)

Note: The numbers in parentheses are p-values.

Turning to the conditional variables, the regression results show that manufacturing productivity

growth has been driven by domestic and foreign R&D intensity, and the distance to the technology

frontier. The estimated coefficients of domestic research intensity are consistently significant in all

the regressions and highly significant in many of the regressions, implying that R&D has permanent

growth effects as predicted by Schumpeterian growth theories. As long as R&D is kept as a constant

proportion of the number of product lines, domestic R&D will keep manufacturing productivity

growth rates constant, ceteris paribus.

The coefficients of the growth in domestic patents are consistently significant in the labor

productivity regressions, however, they are consistently insignificant in the TFP regressions, which

could be due to a positive correlation between the growth in patents and capital stock as discussed in

Section 2. Considering only the labor productivity regressions, the overall regression results are not

consistent with the predictions of semi-endogenous growth theory, even in the regressions where the

coefficients of growth in patents are significant. Semi-endogenous growth theory predicts that a one-

off increase in the level of R&D has only temporary growth effects. However, the permanent

positive growth effects of research intensity ensure that R&D has permanent growth effects. Thus, an

increase in the number of patents issued every period in time permanently increases the productivity

growth rate; however, the productivity effects are higher in the short run than in the long run, a result

that is consistent with the transitional dynamics in the models of Peretto (1998, 1999b).

The foreign knowledge spillover variables give further evidence in favor of Schumpeterian

growth theory. The estimated coefficients of the growth in imports of foreign patents are

insignificant in all of the regressions, while most of the estimated coefficients of foreign R&D

intensity spillovers in the 5-year and the 10-year difference regressions are statistically significant.

They are less significant in the 15-year estimates because the number of observations is small

21

compared to the 5-year and the 10-year estimates. These results point towards permanent growth

effects of R&D spillovers through the channel of imports and show that the choice of trade partners

is important for the benefits derived from trade. These results are consistent with the economy-wide

estimates by Madsen (2008b) and the predictions of the Schumpeterian models of Peretto (2003) in

which an economy that opens up to trade generates a larger and more competitive market in which

firms have access to more diverse technologies, which in turn enhances growth.

The favorable results of Schumpeterian growth theory are consistent with the findings of

Laincz and Peretto (2006), Ha and Howitt (2007) and, particularly, Madsen (2008b), and Madsen et

al. (2009, 2010). Madsen (2008b) and Madsen et al. (2009, 2010) find that economy-wide growth

has been driven by research intensity in the OECD countries since 1870, the UK since 1620 and

India since 1953 and find no evidence for semi-endogenous theory. These studies and this paper,

therefore, show that growth has been governed by the Schumpeterian model throughout the first and

second industrial revolutions in the UK, the transition from Hindu growth rates to spectacular growth

rates in India, and the transition from the post-Malthusian growth regime to modern growth rates in

the OECD countries. The only difference between the findings is that the estimated coefficients of

research intensity are consistently more statistically significant here than in the estimates of Madsen

(2008b) and Madsen et al. (2009, 2010), suggesting that manufacturing productivity advances are

driven more directly by innovations in manufacturing than non-manufacturing or that manufacturing

productivity data are of better quality than non-manufacturing productivity data. The importance of

these results is that Schumpeterian growth theory appears to have general validity and is not limited

to certain sectors of the economy, certain countries and certain stages of development.

The estimated coefficients of the distance to the frontier are significantly positive in the

restricted regressions, however, they are not in the unrestricted regressions. This may reflect a high

degree of correlation between DTF, ln(Pat/Emp)DTF, H(DTF), H and ln(Pat/Emp). Similar results

for economy-wide estimates are found by Madsen (2008b). The significance of DTF shows that off-

frontier countries benefit from the technologies that are developed at the frontier. This result is

consistent with Gerschenkron‟s (1962) hypothesis that off-frontier countries with good institutions

can take advantage of backwardness by adopting the technologies that are developed at the frontier.

The further away a country is from the technology frontier, the lower the costs of innovations. It

follows that the growth potential of a backward country is much larger than for a country close to or

at the technology frontier.

The estimated coefficients of the level of educational attainment are insignificant at the 5%

level in all the regressions, while those of the change in educational attainment are statistically

significant in the 5-year difference estimates, however, they are barely significant in the 10 and 15-

22

year difference estimates. This result is consistent with the observation that educational attainment

has increased almost ten-fold over the period from 1870 to 2006 while the productivity growth rates

have only increased modestly during the same period, as pointed out by Pritchett (2006).

Does this mean that scale effects in human capital have to be abandoned? Not necessarily. In

the human capital driven endogenous growth models, human capital is conceptually broader than

educational attainment. Human capital encompasses educational attainment, the quality of education,

vocational training, on-the-job training, and learning-by-doing, which is enhanced every time a new

product is introduced. Thus, educational attainment may not be an adequate proxy for human capital.

Furthermore, the educational attainment data are not likely to be accurate. The educational

attainment data of Baier et al. (2006), which are used here, are constructed from school enrolment

data going back 50 years before the educational attainment data starts (1870). Since there is very

little information about school enrolment before 1870 for many countries in the Mitchell series, for

instance Mitchell (1983), which is the prime source used by Baier et al. (2006), the educational

attainment data are measured with large errors, particularly before WWI. Thus, the regressions may

not have captured the genuine effects of the level of human capital.

4.4 Financial development, openness and conditional convergence

To investigate further which factors have been responsible for the convergence and to check the

robustness of the results in Tables 5-7 this sub-section extends the models estimated above to allow

for financial development, the interaction between financial development and the distance to frontier

and openness. As discussed in Section 2 financial development plays an important role in the

convergence debate. According to Aghion et al. (2004, 2005) off-frontier countries need a certain

level of financial development to take advantage of their backwardness. The higher are the costs of

defrauding a creditor the more likely it is that the advantage of backwardness exceeds the

disadvantage of backwardness. In addition, the more effective and developed are the institutions, the

higher are the costs of defrauding and the more a country can take advantage of its backwardness.

Following Aghion et al. (2004, 2005) the financial development hypothesis is tested by

adding financial development and the interaction between financial development and the distance to

the frontier to the restricted regression model considered in Tables 5-7:

, (6)

23

where F is an indicator for financial development and OP is openness and is measured as the imports

of goods divided by nominal GDP. The level of human capital, the interaction between DTF and

human capital and the interaction between DTF and research intensity are excluded from the

regressions because they were insignificant in the regressions reported in Tables 5-7 and remained

insignificant when they were added to Eq. (6). The model of Aghion et al. (2004, 2005) predicts that

> 0 and that = 0.10

The model predicts that = 0 because the effect of financial development

on growth vanishes once a country reaches a certain level of financial development.

Financial development is measured as bank assets divided by nominal GDP. This measure is

closely related to the ratio of credit to the private sector to nominal GDP, which is used as the

preferred measure by Aghion et al. (2004, 2005). Bank assets predominantly consist of lending and

investment in assets such as bonds and equity, where lending is by far the most important item on

banks‟ asset side (IMF, International Financial Statistics). Bank assets are used as an indicator of

financial development because statistics on credit to the private sector are not available for most

countries before 1950.

Note that innovations influence productivities through patenting as well as through financial

development in Eq. (6). In the model of Aghion et al. (2004, 2005) financial development affects

productivity growth through the channel of R&D because the access to credit enables entrepreneurs

the go-ahead with positive present-value R&D projects. Since only a subset of innovations are

patented and patents give the same weight to significant and insignificant innovations, the financial

development indicator serves as a useful complement to patents for the innovative activity under the

maintained hypothesis of Aghion et al. (2004, 2005).

Table 8. Estimates of Eq. (6) in 5-, 10- and 15-year intervals

Labor Productivity TFP

5 years 10 years 15 years 5 years 10 years 15 years

-0.092 0.112 0.128 -0.073 0.194 0.211

(0.038) (0.086) (0.080) (0.176) (0.001) (0.024)

-0.008 -0.006 -0.008 -0.003 -0.002 0.011

(0.082) (0.186) (0.195) (0.743) (0.794) (0.640)

0.023 0.006 0.006 0.009 -0.002 -0.006

(0.001) (0.066) (0.192) (0.481) (0.755) (0.519)

0.003 0.001 0.001 0.001 -0.001 0.002

(0.418) (0.676) (0.482) (0.818) (0.619) (0.594)

0.012 0.010 0.013 0.020 0.015 0.033

(0.000) (0.001) (0.000) (0.000) (0.005) (0.000)

10

The prediction of > 0 is reverse of the empirical estimates of Aghion et al. (2004, 2005) because they measure the

distance to the frontier as the minus of the DTF used here.

24

0.004 0.003 0.014 0.006 0.005 0.025

(0.065) (0.049) (0.287) (0.053) (0.110) (0.526)

0.068 0.016 0.084 0.118 0.076 0.244

(0.042) (0.617) (0.024) (0.022) (0.103) (0.008)

0.000 0.005 0.002 -0.006 0.005 0.004

(0.960) (0.250) (0.583) (0.563) (0.371) (0.641)

0.004 0.003 0.004 0.004 0.003 0.007

(0.023) (0.061) (0.069) (0.110) (0.048) (0.063)

0.162 0.053 0.040 0.148 0.010 -0.007

(0.060) (0.100) (0.330) (0.145) (0.856) (0.936)

-0.038 -0.030 -0.038 -0.034 -0.024 -0.031

(0.000) (0.000) (0.000) (0.002) (0.006) (0.205)

0.017 0.014 0.019 0.039 0.033 0.073 (0.028) (0.081) (0.053) (0.006) (0.005) (0.026)

Note. The figures in parenthesis are p-values

The results of regressing Eq. (6) are presented in Table 8. The estimated coefficients of openness are

positive and significant in some regressions while insignificant in others. Furthermore, the estimates

support the financial development hypothesis since the estimated coefficients of the interaction

between financial development and DTF are positive and significant at the 11-percent level in all

cases. The interaction terms become much more significant if DTF is omitted from the estimates (the

results are not shown). Furthermore, consistent with the predictions of the model of Aghion et al.

(2004, 2005) the coefficients of financial development are not significantly different from zero.

Overall, the regressions give strong support for the financial development hypothesis.

The estimated coefficients of the other conditioning variables are largely consistent with the

estimates in the previous sub-section and, the coefficients of domestic research intensity and the

distance to the frontier remain highly significant. Furthermore, the coefficients of the initial

productivity remain insignificant reinforcing the conclusion from the previous section that the

convergence has been driven by the conditioning variables in the model.

5. Concluding remarks

This paper has found strong evidence of manufacturing productivity convergence among the OECD

countries and that the convergence has been driven by research intensity, technology spillovers

through the channel of imports, catching up to the technology frontier through the channel of

financial development and catching up to the frontier independently of financial development.

Focusing on a long historical time-span and using cross-section data, various tests gave evidence of

-convergence as well as -convergence. Further evidence of unconditional -convergence is found

in panel regressions using 5, 10 and 15 year differences. However, initial income was rendered

25

insignificant when growth is conditioned on domestic R&D, technology spillovers, the distance to

the frontier and the interaction between financial development and the distance to the frontier, which

suggests that these variables have been responsible for the convergence. Coupled with good

institutions, the countries with low productivity back in 1870 increased the import of technology

from research intensive countries and took advantage of their backwardness by adapting the

technologies that were developed at the frontier. Financial development played an important role for

countries to take advantage of their backwardness.

The estimation results also gave insight into endogenous growth theories by explaining

manufacturing productivity growth in the industrial countries over the past 137 years. Most

endogenous growth theories suggest that growth is determined by domestic human capital, domestic

R&D and international technology spillovers. However, growth theories have quite different

implications for the functional relationship between growth and domestic as well as foreign R&D,

due to differences in proliferation and scale effects in ideas production. The empirical estimates

showed that the functional relationship between productivity growth, on the one hand, and domestic

and foreign R&D on the other hand, follows the predictions of Schumpeterian growth theory and not

that of semi-endogenous growth theory. It was found that domestic research intensity and foreign

research intensity spillovers through the channel of imports have permanent growth effects, as

predicted by Schumpeterian growth theories. These results are important because they show that

manufacturing productivity growth will remain positive into the future provided that research effort

remains positive. Furthermore, since convergence has been almost completed among OECD

countries, productivity growth rates will converge provided that research intensities converge.

Finally, the regressions showed that backward countries with a developed financial system could

take advantage of their backwardness, increase their innovative activity and speed up their

convergence to the frontier countries.

26

Data appendix.

Manufacturing value-added. After 1960. OECD, National Accounts, Vol II. The following sources

are used before that: Canada. 1870-1926. M C Urquhart, 1988, “Canadian Economic Growth 1870-

1980,” Queens University Discussion Paper No 734. 1926-1960. T Liesner, 1989, One Hundred

Years of Economic Statistics, Oxford: The Economist. USA. B R Mitchell, 1983, International

Historical Statistics: Americas and Australasia, London: Macmillan. Japan. 1870-1884. B R

Mitchell, 1982, International Historical Statistics: Asia and Africa, London: Macmillan. 1885-1940.

K Ohkawa, M Shinchara and L Meissner, 1979, Patterns of Japanese Economic Development: A

Quantitative Appraisal, New Haven: Yale University Press and Liesner, 1989. Australia. W

Vamplew, 1987, Australian Historical Statistics, Broadway, N.S.W: Fairfax, Syme & Weldon

Associates. Belgium. 1870-1900. League of Nations, 1945, Industrialization and Foreign Trade.

1900-1957. C Carbonnelle, 1959, “Recherches sur l‟evolution de la Production en Belgie de 1900 a

1957,” Cahiers Economiques De Bruxelles, 1, 353-377. Denmark. 1870-1920 and 1939-1950. S A

Hansen, 1976, Økonomisk Vækst I Danmark, København: Akademisk Forlag. 1920-1938. OEEC,

1958, Industrial Statistics, 1900-1957, Paris. Finland. R Hjerppe, 1989, The Finnish Economy 1860-

1985, Helsinki: Bank of Finland, Government Printing Centre. France. 1870-1913. B R Mitchell,

1975, European Historical Statistics 1750-1975, London: Macmillan. 1913-1938. OEEC, 1958, op

cit. 1938-1960. Liesner, 1989, op cit. Germany. 1870-1885. W G Hoffmann, F Grumbach and H

Hesse, 1965, Das Wachstum der Deutschen Wirtschaft seit der mitte des 19. Jahrhunderts, Berlin:

Springer-Verlag. 1885-1960. Liesner, 1989, op cit. and OEEC, 1958, op cit. Ireland.1870-1926. C

Mitchell. 1926-1939. OEEC, 1958, op cit. 1939-1960. K A Kennedy, 1971, Productivity and

Industrial Growth: The Irish Experience. Italy. 1870-1885. League of Nations, 1945,

Industrialisation and Foreign Trade. 1885-1960. Liesner, 1989, op cit. The Netherlands. 1870-1913.

J P Smits, E Horlings and J L van Zanden, 2000, Dutch GNP and its Components, 1800-1913,

Groningen, http://www.eco.rug.nl/ggdc/PUB/dutchgnp.pdf. 1913-1950. I Svennilson, 1954, Growth

and Stagnation in the European Economy, Geneva: United Nations Economic Commission for

Europe. 1950-1960. OECD, National Accounts, Vol II. Norway. 1900-1929. OEEC, 1958, op cit.

1929-1960. Statistisk Sentralbyraa, 1968, Nasjonalregnskap, Oslo. Portugal. Industrial Output: C

Bardini, A Carreras and P Lains, 1995, “National Accounts for Italy, Spain and Portugal,”

Scandinavian Economic History Review XLII, 1, 115-146. Spain. L Prados de la Escosura, 2003, El

Progresso Economico De Espana 1850-2000, Madrid: Fundacion BBVA. Sweden. O Krantz and C

A Nilsson, 1975, Swedish National Product 1861-1970, Gleerup: C. W. K. Switzerland. 1870-1913.

Mitchell, 1975, op cit. 1913-1938. Svennelson. UK. 1870-1885. B R Mitchell, 1962, Abstract in

British Historical Statistics, Cambridge: Cambridge University Press. 1885-1960. Liesner, 1989, op

cit.

Manufacturing employment. 1960-2006. OECD, National Accounts, Vol II. 1925-1960 for some

countries: OECD, National Accounts, Vol II, ILO, Yearbook, and United Nations, Statistical

Yearbook. The following sources are used before 1960 or 1929 (note the data are available in ten-

year intervals when P Bairoch (ed.), 1968, “the working population and its structure,” International

historical statistics vol 1, Universite Libre de Bruxelles, Institut de Sociologie, is used. Geometric

interpolation is used between the ten-year intervals): Canada. F H Leacy (ed.), 1983, Historical

Statistics of Canada, Ottawa: Statistics Canada, and Liesner, 1989, op cit. USA. 1870-1900. Bairoch,

1968, op cit. 1900-1960. Liesner, 1989, op cit. Japan. Liesner, 1989, op cit. and Ohkawa, Kazushi,

1957, The Growth Rate of the Japanese Economy since 1878, Kinokuniya Bookstore. Belgium.

Bairoch, 1968, op cit. Denmark. 1930-60. H C Johansen, 1985, Dansk Historisk Statistik 1814-1980,

København: Gyldendal. Finland. R Hjerppe, 1989, The Finnish Economy 1860-1985, Helsinki: Bank

of Finland, Government Printing Centre. France. D Rouzet, 2004-2005, “L‟evolution de salaries et

de la rente fonciere en France (1450-194),” DEA Analyse et Politiques Economiques. 1930-1939 and

P Villa, 1993, Une Analyse Macroeconomique De La France Au Xxe Siecle, Paris: CNRS Editions.

27

Germany. Hoffmann et al., 1965, op cit. Ireland. 1870-1926. Bairoch, 1968, op cit. 1926-1960.

Kennedy. Italy. G Fua, 1965, Notes on Italian Economic Growth 1861-1964, Milano: M Pavcis. The

Netherlands. 1870-1913. Smits et al., 2000, op cit. 1920. Bairoch, 1968, op cit. Interpolated 1913-

1920 and 1920-1929. Norway. 1870-1929. Bairoch, 1968, op cit. Portugal. N Valerio, 2001,

Portuguese Historical Statistics INE: Lisbon. Spain. Instituto de Estudies Fiscales, 1978, “Datos

Basicos Para La Historia Financiera De Espana (1850-1975)” Madrid: Ministoio de Hacienda.

Sweden. K G Jungenfelt, 1966, Lønandelen och den ekonomiska utvecklingen, Stockholm: Almqvist

& Wiksell. UK. S Broadberry and C Burhop, 2007, “Comparative Productivity in British and

German Manufacturing before World War II: Reconciling Direct Benchmark Estimates and Time-

Series Evidence,” Journal of Economic History, 67, 315-349, and Liesner, 1989, op cit.

ILO employment data. ILO, Yearbook.

Purchasing power parities. Since data on manufacturing PPP is not available, PPPs for GDP are

used instead. OECD, 2005, Purchasing Power Parities and Real Expenditures, Vol. 1, Paris. The

PPPs in 1980 and 1990 are from http:/www.oecd.org/std/ppp. OECD, accessed in 2008, Purchasing

Power Parities (PPPs) for OECD Countries since 1980, Paris.

Human capital. Baier, S.L., Dwyer, G.P. and R Tamura, 2006, “How Important are Capital and

Total Factor Productivity for Economic Growth?” Economic Inquiry, 44, 23-49. The data are

available online at www.jerrydwyer.com.

Openness. Nominal imports divided by nominal GDP. See J B Madsen, (2009). “Trade Barriers, Openness and Economic Growth,” Southern Economic Journal, 76, 397-418.

Economy-wide nominal GDP. See Madsen (2009) op cit.

Annual hours worked per employee. See Madsen (2009) op cit.

Manufacturing capital stock. Estimated from gross investment using the perpetual inventory

method and an eight percent depreciation rate. The initial capital stock is estimated as the average

investment over the first five years divided by the depreciation rate plus the average annual

geometric growth rate in investment from the first to the last observation. Over the period from 1950,

or later, to 2006 the following data sources are used for all countries:

OECD, National Accounts, Vol II, Paris, and OECD, Flows and Stocks of Fixed Capital, Paris. The

following sources have been used for earlier data for each individual country. Canada. Leacy, 1983,

op cit. USA. S B Carter, S S Gartner, M R Haines, A L Olmstead, R Sutch and G Wright, 2006,

Historical Statistics of the United States Millennial Edition, Cambridge: Cambridge University

Press. Japan. K Ohkawa et al., 1979, op cit. Belgium. M van Meerten, 2003, Capital Formation in

Belgium, 1900-1995, Leuven: Leuven University Press. Denmark. Hansen, 1974, op cit., non-

residential non-agricultural investments. They are nominal and are deflated by the overall investment

deflator. Finland. M Lindmark and P Vikström, 2003, “Growth and structural change in Sweden and

Finland 1870-1990: a story of convergence,” Scandinavian Economic History Review, 51, 46-74.

France. Kindly provided by B van Ark, Groningen Growth and Development Centre, 2005.

Germany. W Kirner, 1968, Zeitreihen fur das Anlagevermogen der Wirtschaftsbereiche in der

Bundesreplublik Deutschland, Deutsches Institut fur Wirtschaftsforschnung, Berlin: Duncker &

28

Humbolt. The data are adjusted for war damage in the source. Ireland. K A Kennedy, 1971,

Productivity and Industrial Growth: The Irish Experience, Clarendon Press, Oxford. Netherlands.

Kindly provided by B van Ark, Groningen Growth and Development Centre. Norway. Statistikbyrå,

(1968), National Accounts, 1865-1960, Oslo: Statistikbyrå, table 34. Sweden. Ö Johansson, 1967,

The Gross Domestic Product of Sweden and its Composition 1861-1955, Stockholm: Almqvist and

Wiksell, table 46. UK. C H Feinstein: "Statistical Tables of National Income, Expenditure and

Output of the U.K. 1855-1965", Cambridge: Cambridge University Press.

Bank assets. 1948-2006 for all countries. IMF, International Financial Statistics. Before 1948.

Canada. R W Goldsmith, 1969, Financial Structure and Development, New Haven: Yale University

Press. USA. Historical Statistics, op cit. Japan. Goldsmith, 1969, op cit. Australia. S J Butlin, A R

Hall and R C White, 1971, Australian Banking and Monetary Statistics 1817-1945, 4B ,Australian

Banking and Monetary Statistics 1945-1970, RBA Occasional Papers 4A. New Zealand. 1880-1963.

Goldsmith, 1969, op cit. Belgium. Goldsmith, 1969, op cit. Denmark. 1875-2005. K Abildgren,

2006, "Monetary Trends and Business Cycles in Denmark since 1875,” Denmark’s National Bank

Working Papers, and H C Johansen, 1985, op cit. Finland. Bank assets are proxied by M1 before

1948. T. Haavisto, (1992), Money and Economic Activity in Finland 1866-1985, Lund: Lund

Economic Studies. France. 1860-1963. Goldsmith, 1969, op cit. Germany. Goldsmith, 1969, op cit.

Ireland. 1880-1930, D K Sheppard, 1971, The Growth and Role of UK Financial Institutions 1880-

1962. London, Methuen. 1931-1980. B R Mitchell, 1988, British Historical Statistics, Cambridge:

Cambridge University Press. Italy. Goldsmith, 1969, op cit. Netherlands. Goldsmith, 1969, op cit.

Norway. Goldsmith, 1969, op cit. Portugal. N Valerio, 2001, op cit. Spain. Goldsmith, 1969, op cit.

And Estadisticas hisotricas de Espana Vol. II.

Sweden. 1880-1963. W R Goldsmith, 1969, op cit. Switzerland. 1826-1906. CHRONOS Historical

Statistics of Switzerland, table O.12. 1907-1963. W R Goldsmith, 1969, op cit. UK. 1860-1879. W R

Goldsmith, 1969, op cit. 1880-1966. D K Sheppard, 1971, op cit.

Total employment. 1960-2006 OECD, National Accounts, Vol 2. Before 1960 the following

sources are used: The algorithm which is suggested by V Gomez and A Maravall, 1994, op cit. is

used to interpolate between the benchmark years as indicated for the individual countries. Canada.

1921-1959. F H Leacy (ed.), 1983, Historical Statistics of Canada, Ottawa: Statistics Canada. 1870,

1890, and 1913, and A Maddison, 1991, Dynamic Forces in Capitalist Development, Oxford: Oxford

University Press. The US. 1900-1949. T Liesner, 1989, op cit. 1870, 1890, and 1893: A Maddison,

1991, op cit. Japan. K Ohkawa, M Shinchara and L Meissner, 1979, op cit. Australia. 1901-1949. M

W Butlin, 1977, A Preliminary Annual Database 1900/01 to 1973/74, Research Discussion Paper

7701, Sydney: Reserve Bank of Australia. A Maddison, 1991, op cit. Denmark. 1870-1949. S A

Hansen, 1976, op cit. Finland. 1870-1959. R Hjerppe, 1989, op cit. Germany. 1870-1872, 1874-

1914, 1924-1940, and 1949. W G Hoffmann, F Grumbach and H Hesse, 1965, Das Wachstum der

Deutschen Wirtschaft seit der mitte des 19. Jahrhunderts, Berlin: Springer-Verlag. Italy. 1901-1949.

C Clark, 1957, op cit. 1870, and 1990. A Maddison, 1991, op cit. Netherlands. Central Bureau voor

de Statistiek, 2001, Tweehondred Jaar Statistiek in Tijdreeksen, 1800-1999, Centraal Bureau voor de

Statistiek, Voorburg. Norway. 1903-1919. P Flora, F Kraus and P W Phenning, 1987, State,

Economy, and Society in Western Europe 1815-1975, London: Macmillan. 1920-1949. C Clark,

29

1957, op cit. 1870, and 1890. A Maddison, 1991, op cit. Sweden. Ö Johansson, 1967, op cit. UK. C

Clark, 1957, op cit.

30

Appendix. Results from using other estimators

Fixed Effects OLS

5 years Full Model Restricted

Model Full Model

Restricted

Model

Y/L TFP Y/L TFP Y/L TFP Y/L TFP

-0.010 -0.007 -0.005 0.006 -0.001 -0.005 -0.003 -0.002

(0.238) (0.603) (0.266) (0.372) (0.840) (0.737) (0.586) (0.860)

0.017 0.017 0.017 0.022 0.014 0.039 0.012 0.040

(0.015) (0.194) (0.010) (0.085) (0.067) (0.005) (0.100) (0.003)

0.006 0.003 0.006 0.003 0.008 0.005 0.009 0.005

(0.102) (0.530) (0.100) (0.531) (0.056) (0.425) (0.049) (0.420)

0.004 0.020 0.005 0.015 0.002 0.008 0.004 0.007

(0.215) (0.001) (0.033) (0.002) (0.220) (0.002) (0.005) (0.007)

0.004 0.006 0.005 0.007 0.001 0.001 0.001 0.001

(0.048) (0.041) (0.008) (0.008) (0.110) (0.582) (0.065) (0.487)

0.149 0.265 0.147 0.225 0.095 0.155 0.124 0.155

(0.093) (0.030) (0.087) (0.054) (0.175) (0.097) (0.069) (0.070)

-0.033 -0.028 -0.033 -0.025 -0.028 -0.028 -0.029 -0.028

(0.000) (0.007) (0.000) (0.010) (0.002) (0.043) (0.001) (0.036)

0.015 0.029 0.022 0.057 0.001 0.003 0.016 0.023

(0.479) (0.353) (0.001) (0.000) (0.940) (0.916) (0.049) (0.118)

0.004 0.011

0.014 0.012

(0.821) (0.573)

(0.270) (0.469)

0.001 -0.006

0.001 -0.002

(0.575) (0.210)

(0.358) (0.385)

0.017 0.039

-0.005 0.006

(0.465) (0.262)

(0.659) (0.751)

C 0.105 0.046 0.085 -0.016 0.051 0.061 0.058 0.044 (0.059) (0.604) (0.055) (0.814) (0.344) (0.570) (0.246) (0.634)

Fixed Effects OLS

10 years Full Model Restricted

Model Full Model

Restricted

Model

Y/L TFP Y/L TFP Y/L TFP Y/L TFP

-0.008 -0.006 -0.007 0.001 -0.008 -0.009 -0.003 -0.006

(0.238) (0.626) (0.059) (0.828) (0.238) (0.369) (0.349) (0.340)

0.007 0.001 0.006 0.004 0.007 0.012 0.004 0.014

(0.036) (0.876) (0.035) (0.423) (0.036) (0.043) (0.232) (0.015)

0.001 0.001 0.001 0.000 0.001 0.001 0.003 0.001

(0.714) (0.823) (0.665) (0.888) (0.714) (0.768) (0.315) (0.784)

0.004 0.015 0.005 0.011 0.004 0.006 0.003 0.004

(0.131) (0.001) (0.009) (0.007) (0.131) (0.007) (0.006) (0.032)

0.003 0.004 0.004 0.005 0.003 0.000 0.001 0.000

(0.039) (0.121) (0.009) (0.018) (0.039) (0.950) (0.058) (0.782)

0.055 0.078 0.054 0.063 0.055 0.053 0.051 0.051

(0.134) (0.121) (0.114) (0.180) (0.134) (0.210) (0.086) (0.192)

-0.032 -0.025 -0.032 -0.025 -0.032 -0.030 -0.028 -0.030

(0.000) (0.002) (0.000) (0.002) (0.000) (0.002) (0.000) (0.001)

0.011 0.004 0.011 0.035 0.011 -0.010 0.012 0.012

(0.533) (0.864) (0.041) (0.000) (0.533) (0.641) (0.004) (0.093)

0.000 0.018

0.000 0.014

(0.986) (0.301)

(0.986) (0.380)

0.001 -0.006

0.001 -0.003

(0.526) (0.091)

(0.526) (0.164)

0.006 0.020

0.006 0.009

(0.753) (0.465)

(0.753) (0.629)

C 0.110 0.062 0.106 0.034 0.110 0.102 0.065 0.085 (0.015) (0.400) (0.003) (0.544) (0.015) (0.162) (0.065) (0.136)

31

Fixed Effects OLS

15 years Full Model Restricted

Model Full Model

Restricted

Model

Y/L TFP Y/L TFP Y/L TFP Y/L TFP

-0.019 0.009 -0.005 0.034 -0.004 -0.014 -0.005 0.002

(0.062) (0.815) (0.321) (0.045) (0.519) (0.587) (0.330) (0.874)

0.007 0.002 0.007 0.006 0.005 0.011 0.004 0.013

(0.033) (0.821) (0.015) (0.493) (0.106) (0.111) (0.144) (0.055)

0.003 0.006 0.003 0.005 0.004 0.006 0.004 0.006

(0.096) (0.089) (0.040) (0.094) (0.059) (0.108) (0.044) (0.112)

0.006 0.031 0.009 0.025 0.002 0.013 0.005 0.009

(0.055) (0.000) (0.000) (0.001) (0.230) (0.002) (0.002) (0.003)

0.021 0.040 0.013 0.043 0.014 0.016 0.013 0.020

(0.089) (0.159) (0.264) (0.124) (0.054) (0.305) (0.064) (0.198)

0.104 0.163 0.081 0.139 0.048 0.101 0.053 0.082

(0.007) (0.051) (0.016) (0.053) (0.093) (0.138) (0.047) (0.136)

-0.035 -0.012 -0.033 -0.008 -0.032 -0.043 -0.033 -0.040

(0.000) (0.623) (0.000) (0.674) (0.000) (0.040) (0.000) (0.025)

0.026 0.011 0.020 0.097 0.008 -0.010 0.013 0.036

(0.226) (0.815) (0.004) (0.000) (0.625) (0.783) (0.017) (0.038)

-0.010 0.045

0.007 0.022

(0.552) (0.151)

(0.583) (0.382)

0.003 -0.015

0.002 -0.007

(0.192) (0.047)

(0.226) (0.090)

0.046 0.061

0.002 0.034

(0.085) (0.380)

(0.869) (0.352)

C 0.116 -0.167 0.062 -0.300 0.059 0.105 0.067 0.011 (0.059) (0.515) (0.187) (0.063) (0.251) (0.578) (0.149) (0.933)

32

References

Aghion, P., and P. Howitt, Endogenous Growth (Cambridge, MA: MIT Press, 1998).

____, “Growth with Quality-Improving Innovations: An Integrated Framework” (pp. 67-110), in Philippe Aghion and

Steven N. Durlauf (Eds.), Handbook of Economic Growth, Volume 1A (Amsterdam: North-Holland, 2005).

Aghion, P., P. Howitt, and D. Mayer-Foulkes, “The Effect of Financial Development on Convergence: Theory and

Evidence,” NBER working paper no. 10358 (2004).

____, “The Effect of Financial Development on Convergence: Theory and Evidence,” Quarterly Journal of Economics

120:1 (2005), 173-221.

Aghion, P., C. Meghir, and J. Vandenbussche, “Growth, Distance to Frontier and the Composition of Human Capital,”

Journal of Economic Growth 11:2 (2006), 97-127.

Baier, Scott L, Gerald P Dwyer, and Robert Tamura, “How Important are Capital and Total Factor Productivity for

Economic Growth?” Economic Inquiry 44:1 (2006), 23-49.

Baumol, W.J., “Productivity Growth, Convergence and Welfare: What the Long-Run Data Show,” American Economic

Review 76:5 (1986), 1072-1085.

Baumol, W.J., and E.N. Wolff, “Productivity Growth, Convergence and Welfare: Reply,” American Economic Review

78:5 (1988), 1155-1159.

Ben-David, D., “Equalizing Exchange: Trade Liberalization and Income Convergence,” Quarterly Journal of Economics

108:3 (1993), 653–679.

Bernard, A.B., and C.I. Jones, “Productivity Across Industries and Countries: Time Series Theory and Evidence,” Review

of Economics and Statistics 78:1 (1996a), 135-146.

____, “Comparing Apples to Oranges: Productivity Convergence and Measurement Across Industries and Countries,”

American Economic Review 86:5 (1996b), 1216-1238.

Broadberry, S.N., “Manufacturing and the Convergence Hypothesis; What the Long-Run Data Show,” Journal of

Economic History 53:4 (1993), 772-795.

Bruno, Michael, and Jeffrey Sachs, Economics of Worldwide Stagflation (Cambridge: Harvard University Press, 1985).

Carree, M., and L. Klomp, “Testing the Convergence Hypothesis: A Comment,” Review of Economics and Statistics 79:4

(1997), 683-686.

Carree, M.A., L. Klomp, and A.R. Thurik, “Productivity Convergence in OECD Manufacturing Industries,” Economic

Letters 66:3 (2000), 337-345.

Clark, G., and R.C. Feenstra, “Technology and the Great Divergence” (pp. 277-322), in Michael D. Bordo, Alan Taylor

and Jeffrey Williamson (Eds.), Globalization in Historical Perspective (Chicago: University of Chicago Press,

2003).

Coe, D.T., and E. Helpman, “International R&D spillovers,” European Economic Review 39:5 (1995), 859-897.

De Long, J.B., “Productivity Growth, Convergence, and Welfare: Comment,” American Economic Review 78:5 (1988),

1138-1154.

Dinopoulos, E., and P. Thompson, “Schumpeterian Growth without Scale Effects," Journal of Economic Growth 3:4

(1998), 313-335.

Di Liberto, A., F. Pigliaru, and R. Mura, “How to Measure the Unobservable: A Panel Technique for the Analysis of TFP

Convergence,” Oxford Economic Papers 60:2 (2008), 343-368.

Dollar, D., and E.N. Wolff, “Convergence of Industry Labor Productivity among Advanced Economies, 1963-1982,”

Review of Economics and Statistics 70:4 (1988), 549-558.

Dowrick, S., and N. Gemmell, “Industrialisation, Catching Up and Economic Growth: A Comparative Study across the

World's Capitalist Economies,” Economic Journal 101:405 (1991), 263-275.

Dowrick, S., and D.T. Nguyen, “OECD Comparative Economic Growth 1950-85: Catch-Up and Convergence,”

American Economic Review 79:5 (1989), 1010-1030.

Eaton, J., and S. Kortum, “International Technology Diffusion: Theory and Measurement,” International Economic

Review 40:3 (1999), 537-570.

Gerschenkron, A., Economic Backwardness in Historical Perspective (Cambridge: Belknap Press, 1962).

Gollin, D., “Getting Income Shares Right,” Journal of Political Economy 110:2 (2002), 458-474.

Griliches, Z., “Issues in Assessing the Contribution of Research and Development to Productivity Growth,” Bell Journal

of Economics 10:1 (1979), 92-116.

Griffith, R., S. Redding, and J.V. Reenen, “R&D and Absorptive Technology: Theory and Empirical Evidence,”

Scandinavian Journal of Economics 105:1 (2003), 99-118.

____ , “Mapping the Two Faces of R&D: Productivity Growth in a Panel of OECD Industries,” Review of Economics

and Statistics 86:4 (2004), 883-895.

Ha, J., and P. Howitt, “Accounting for Trends in Productivity and R&D: A Schumpeterian Critique of Semi-Endogenous

Growth Theory,” Journal of Money Credit and Banking 39:4 (2007), 733-774.

Hall, Robert E, “The Relation between Price and Marginal Cost in U.S. Industry,” Journal of Political Economy 96:5

(1988), 921-947.

33

Hobday, M., “Innovation in Asian Industrialization: A Gerschenkronian Perspective,” Oxford Development Studies 31:3

(2003), 293-314.

Howitt, P., “Steady Endogenous Growth with Population and R&D Growing,” Journal of Political Economy 107:4

(1999), 715-730.

____, “Endogenous Growth and Cross-Country Income Differences,” American Economic Review 90:4 (2000), 829-846.

Howitt, P., and D. Mayer-Foulkes, “R&D, Implementation and Stagnation: A Schumpeterian Theory of Convergence

Clubs,” Journal of Money, Credit and Banking 37:1 (2005), 147-177.

Jaggers, K., and M.G. Marshall, “Polity IV Project: Political Regime Characteristics and Transitions, 1800-2006,” Center

for Systemic Peace, University of Maryland, Available at www.systemicpeace.org/polity (2007).

Jones, C.I., “R&D Based Models of Economic Growth,” Journal of Political Economy 103:4 (1995), 759-784.

____, “Convergence Revisited,” Journal of Economic Growth 2:2 (1997), 131-153.

____, “Sources of U.S. Economic Growth in a World of Ideas,” American Economic Review 92:1 (2002), 220-239.

Kelly, M., “On Endogenous Growth with Productivity Shocks,” Journal of Monetary Economics 30:1 (1992), 47-56.

Kiviet, J.F., “On Bias, Inconsistency, and Efficiency of Various Estimators in Dynamic Panel Data Models,” Journal of

Econometrics 68:1 (1995), 53-78.

Kortum, S., “Research, Patenting, and Technological Change,” Econometrica 65:6 (1997), 1389-1419.

Laincz, C., and P.F. Peretto, “Scale Effects in Endogenous Growth Theory: An Error of Aggregation not Specification,”

Journal of Economic Growth 11:3 (2006), 263-288.

Lichtenberg, F.R., “Testing the Convergence Hypothesis,” Review of Economics and Statistics 76:3 (1994), 576-579.

Lichtenberg, F.R., and B. Van Pottelsberghe de la Potterie, “International R&D Spillovers: A Comment,” European

Economic Review 42:8 (1998), 1483-1491.

Lucas, R., Jr., “On the Mechanisms of Economic Development,” Journal of Monetary Economics 22:1 (1988), 3-42.

Madsen, J.B., “Technology Spillover through Trade and TFP Convergence: 135 Years of Evidence for the OECD

Countries,” Journal of International Economics 72:2 (2007), 464-480.

____, “Economic Growth, TFP Convergence and the World Export of Ideas: A Century of Evidence,” Scandinavian

Journal of Economics 110:1 (2008a), 145-167.

____, “Semi-Endogenous versus Schumpeterian Growth Models: Testing the Knowledge Production Function using

International Data,” Journal of Economic Growth 13:1 (2008b), 1-26.

____, “Trade Barriers, Openness, and Economic Growth,” Southern Economic Journal 76:2 (2009), 397-418.

Madsen, J. B., J. B. Ang, and R. Banerjee, “Second-Generation Innovation-Based Growth Models and the British

Industrial Revolution,” Paper presented at the 2009 Australasian Macroeconomic Workshop, Deakin University,

(2009).

Madsen, J. B., J. B. Ang, and S. Saxena, “The Indian Growth Miracle and Endogenous Growth,” Journal of Development

Economics (2010), forthcoming.

Mankiw, N.G., D. Romer, and D. Weil, “A Contribution to the Empirics of Economic Growth,” Quarterly Journal of

Economics 107:2 (1992), 407-437.

Mitchell, B.R., International Historical Statistics: Americas and Australasia, (Macmillan, London, 1983).

Nelson, R.R., and E.S. Phelps, “Investment in Humans, Technological Diffusion, and Economic Growth,” The American

Economic Review 56:1/2 (1966), 69-75.

Nelson, R.R., and G. Wright, “The Rise and fall of American Technological Leadership: The Postwar Era in Historical

Perspective,” Journal of Economic Literature 30:4 (1992), 1931-1964.

OECD, Economic Survey for Portugal, OECD Policy Brief, Available at http://www.oecd.org/publications/Pol_brief

(2004).

Peretto, P. F., “Sunk Costs, Market Structure, and Growth,” International Economic Review 37:4 (1996), 895-923.

____, “Technological Change and Population Growth,” Journal of Economic Growth 3:4 (1998), 283-311.

____, “Industrial Development, Technological Change, and Long-Run Growth,” Journal of Development Economics

59:2 (1999a), 389-417.

____, “Cost Reduction, Entry, and the Interdependence of Market Structure and Economic Growth,” Journal of

Monetary Economics 43:1 (1999b), 173-195.

____, “Endogenous Market Structure and the Growth and Welfare Effects of Economic Integration,” Journal

International Economics 60:1 (2003), 177-201.

Peretto, P.F., and S. Smulders, “Technological Distance, Growth and Scale Effects,” Economic Journal 112:481 (2002),

603-624.

Peretto, P.F., and J.J. Seater, “Factor-Eliminating Technical Change,” Economic Research Initiatives at Duke (ERID),

Working Paper no. 17 (2008).

Pritchett, L., “Does Learning to Add up Add Up? The Returns to Schooling in Aggregate Data” (pp.635-695), in Eric

Hanushek and Finis Welch (Eds.), Handbook of the Economics of Education, Volume 1 (The Netherlands: North-

Holland, 2006)

Romer, P.M., “Endogenous Technological Change,” Journal of Political Economy 98:5 (1990), S71-102.

Sala-i-Martin, X.X., “The Classical Approach to Convergence Analysis,” Economic Journal 106:437 (1996), 1019-1036.

34

Sørensen, A., “Comparing Apples to Oranges: Productivity Convergence and Measurement across Industries and

Countries: Comment,” American Economic Review 91:4 (2001), 1160-1167.

Wolf, H.C., “Growth Convergence Reconsidered,” Review of World Economics 130:4 (1994), 747-759.

Wolff, E.N., “Capital Formation and Productivity Convergence over the Long Run,” American Economic Review 81:3

(1991), 565-579.